[
https://issues.apache.org/jira/browse/TS-1247?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Alan M. Carroll reassigned TS-1247:
-----------------------------------
Assignee: Alan M. Carroll
> Data from a response transform in a plugin causes excess memory usage
> ---------------------------------------------------------------------
>
> Key: TS-1247
> URL: https://issues.apache.org/jira/browse/TS-1247
> Project: Traffic Server
> Issue Type: Bug
> Components: HTTP
> Affects Versions: 3.0.4
> Reporter: William Bardwell
> Assignee: Alan M. Carroll
> Priority: Minor
> Fix For: 3.3.3
>
>
> If you have a plugin implementing a response transform and the data sending
> to the UA is slow TS will use huge amounts of memory for larger files. (e.g.
> 16G Vsize for a 50M file.)
> This seems to be several problems combined, first, the MIOBuffer used for the
> chunking tunnel uses 2M buffers, and when handling data from a transform it
> ends up putting just the chunk markers in a buffer, then gets a buffer from
> the transform, then a new 2M buffer...Once I fix that, 50M files are ok, but
> 1G files still exhaust memory as they end up all in memory because there is
> no flow-control in some of the VConnections and such between a transform and
> the chunking tunnel. I did a lame patch that I need to revise that fixed the
> problem by making TransformTerminus not send data if the destination that it
> is writing to has too much data in it, plus I made my plugin check how much
> data the destination for the transform has and don't write to it then either.
> But it seems like I need to find the other side of both of those, and
> suppress WRITE_READY (and maybe IMMEDIATE) event sending when they are full,
> but I haven't been able to figure out what the other side is for those yet.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
